HYBRID ARTIFICIAL INTELLIGENCE METHODS FOR THE RELIABILITY ANALYSIS OF MULTI-STOREY FRAME STRUCTURES UNDER VARIOUS LOADING CONDITIONS

Abstract

When civil engineering structures are exposed to unpredictable loading circumstances, structural reliability evaluation is crucial for assuring their safety and performance. When testing how structures react to different random variables, the traditional numerical methods that rely on running finite element simulations again and again are computationally intensive. A framework for dependability analysis of multistorey frame structures exposed to variable lateral and combined loading conditions is proposed in this research, which is based on a mix of artificial intelligence techniques. Using SAP2000, numerical datasets were generated for a two-span, six-story plane frame structure that takes into account lateral stresses, gravity loads, and fluctuations in modulus of elasticity as unknown factors. By combining the results of the structure response data generation process with the optimization algorithms of the Random Forest (RF), Whale (WOA), and Sparrow Search (SSA), a hybrid machine learning model was created. In order to forecast the crucial node's lateral displacement under various loading situations, the models that were constructed were trained and evaluated. Utilizing statistical indicators such as NashSutcliffe efficiency (NS), adjusted R², performance index (PI), variance account factor (VAF), Legate and McCabe index (LMI), Willmott index (WI), root mean square error (RMSE), mean absolute error (MAE), weighted mean absolute percentage error (WMAPE), and scatter index (SI), the suggested models' performance was assessed. In addition, the models' reliability performance was evaluated using the FOSM approach, which measures the reliability index (Η) and the probability of failure (Pf). According to the findings, each of the suggested hybrid models managed to get a remarkable level of prediction accuracy. The RF-WOA model showed better reliability performance and generalizability for structures that were loaded floor-wise laterally. During the testing and training stages, RF-DOA obtained the best prediction accuracy when subjected to a combination of gravity and lateral loads. The results of the reliability study show that hybrid AI models are a good substitute for computationally heavy structural evaluation methods and can accurately forecast the reaction of structures.